File size: 68,541 Bytes
07423df |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 |
import glob
import logging
import os
import shutil
import time
import zipfile
from pathlib import Path
from typing import Callable, List, Optional, Set
import accelerate
import einops
import huggingface_hub
import numpy as np
import pandas as pd
import torch
import transformers
import yaml
from h2o_wave import Q, data, ui
from sqlitedict import SqliteDict
from llm_studio.app_utils.config import default_cfg
from llm_studio.app_utils.hugging_face_utils import (
get_model_card,
publish_model_to_hugging_face,
)
from llm_studio.app_utils.sections.chat import chat_tab, load_cfg_model_tokenizer
from llm_studio.app_utils.sections.common import clean_dashboard
from llm_studio.app_utils.utils import (
add_model_type,
flatten_dict,
get_cfg_list_items,
get_data_dir,
get_download_link,
get_experiment_status,
get_experiments,
get_model_types,
get_problem_categories,
get_problem_types,
get_ui_elements,
get_unique_name,
hf_repo_friendly_name,
parse_ui_elements,
remove_model_type,
set_env,
start_experiment,
)
from llm_studio.app_utils.wave_utils import busy_dialog, ui_table_from_df, wave_theme
from llm_studio.python_configs.cfg_checks import check_config_for_errors
from llm_studio.src.datasets.text_utils import get_tokenizer
from llm_studio.src.tooltips import tooltips
from llm_studio.src.utils.config_utils import (
NON_GENERATION_PROBLEM_TYPES,
load_config_py,
load_config_yaml,
save_config_yaml,
)
from llm_studio.src.utils.exceptions import LLMResourceException
from llm_studio.src.utils.export_utils import (
check_available_space,
get_artifact_path_path,
get_logs_path,
get_model_path,
get_predictions_path,
save_logs,
save_prediction_outputs,
)
from llm_studio.src.utils.logging_utils import write_flag
from llm_studio.src.utils.modeling_utils import unwrap_model
from llm_studio.src.utils.plot_utils import PLOT_ENCODINGS
from llm_studio.src.utils.utils import add_file_to_zip, kill_child_processes
logger = logging.getLogger(__name__)
async def experiment_start(q: Q) -> None:
"""Display experiment start cards."""
await clean_dashboard(q, mode="experiment_start", exclude=["experiment/start"])
q.client["nav/active"] = "experiment/start"
show_update_warnings = True
is_create_experiment = False
# reset certain configs if new experiment start session
if (
q.args.__wave_submission_name__ == "experiment/start"
or q.args.__wave_submission_name__ == "experiment/start_experiment"
or q.args.__wave_submission_name__ == "dataset/newexperiment"
or q.args.__wave_submission_name__ == "dataset/newexperiment/from_current"
or q.args.__wave_submission_name__ == "experiment/list/new"
):
q.client["experiment/start/cfg_experiment_prev"] = None
q.client["experiment/start/cfg_file_prev"] = None
q.client["experiment/start/prev_dataset"] = None
q.client["experiment/start/cfg_sub"] = None
show_update_warnings = False
is_create_experiment = True
# get all the datasets available
df_datasets = q.client.app_db.get_datasets_df()
# Hide inference only datasets
df_datasets = df_datasets.loc[df_datasets["train_rows"].notna()]
if (
not q.client["experiment/start/dataset"]
or q.client["experiment/start/dataset"] not in df_datasets.id.astype(str).values
):
if len(df_datasets) >= 1:
q.client["experiment/start/dataset"] = str(df_datasets["id"].iloc[-1])
else:
q.client["experiment/start/dataset"] = "1"
warning_message = "Experiment settings might be updated after changing {}"
items = [
ui.separator(name="general_expander", label="General settings"),
ui.dropdown(
name="experiment/start/dataset",
label="Dataset",
required=True,
value=q.client["experiment/start/dataset"],
choices=[
ui.choice(str(row["id"]), str(row["name"]))
for _, row in df_datasets.iterrows()
],
trigger=True,
tooltip=tooltips["experiments_dataset"],
),
]
if (
show_update_warnings
and q.client["experiment/start/dataset_prev"]
!= q.client["experiment/start/dataset"]
):
items += [
ui.message_bar(type="warning", text=warning_message.format("Dataset"))
]
show_update_warnings = False
if (
q.client["experiment/start/cfg_file"] is None
or q.client["experiment/start/dataset_prev"]
!= q.client["experiment/start/dataset"]
) and q.client["experiment/start/cfg_category"] != "experiment":
dataset = q.client.app_db.get_dataset(q.client["experiment/start/dataset"])
if dataset is not None:
problem_type = dataset.config_file.replace(dataset.path + "/", "").replace(
".yaml", ""
)
else:
problem_type = default_cfg.cfg_file
q.client["experiment/start/cfg_file"] = problem_type
q.client["experiment/start/cfg_category"] = problem_type.split("_")[0]
if q.client["experiment/start/cfg_category"] == "experiment":
q.client["experiment/start/cfg_file"] = "experiment"
# get all experiments
df_experiments = get_experiments(q, mode="train")
# get all problem category choices
choices_problem_categories = [
ui.choice(name, label) for name, label in get_problem_categories()
]
if len(df_experiments["id"]) > 0:
choices_problem_categories += [ui.choice("experiment", "From Experiment")]
# set default value of problem type if no match to category
if (
q.client["experiment/start/cfg_category"]
not in q.client["experiment/start/cfg_file"]
):
if q.client["experiment/start/cfg_category"] != "experiment":
q.client["experiment/start/cfg_file"] = get_problem_types(
category=q.client["experiment/start/cfg_category"]
)[0][0]
# get all problem type choices
choices_problem_types = [
ui.choice(name, label)
for name, label in get_problem_types(
category=q.client["experiment/start/cfg_category"]
)
]
# remove model type if present in cfg file name here
q.client["experiment/start/cfg_file"] = remove_model_type(
q.client["experiment/start/cfg_file"]
)
if len(df_experiments["id"]) > 0:
if q.client["experiment/start/cfg_experiment"] is None:
q.client["experiment/start/cfg_experiment"] = str(
df_experiments["id"].iloc[0]
)
# Default pretrained from the previous experiment to False
if (
q.client["experiment/start/cfg_experiment_pretrained"] is None
or is_create_experiment
):
q.client["experiment/start/cfg_experiment_pretrained"] = False
if q.client["experiment/start/cfg_category"] != "experiment":
items += [
ui.dropdown(
name="experiment/start/cfg_file",
label="Problem Type",
required=True,
choices=choices_problem_types,
value=q.client["experiment/start/cfg_file"],
trigger=True,
tooltip=tooltips["experiments_problem_type"],
)
]
model_types = get_model_types(q.client["experiment/start/cfg_file"])
if len(model_types) > 0:
choices = [ui.choice(name, label) for name, label in model_types]
if q.client["experiment/start/cfg_sub"] in [None, ""]:
q.client["experiment/start/cfg_sub"] = model_types[0][0]
items += [
ui.dropdown(
name="experiment/start/cfg_sub",
label="Model Type",
required=True,
choices=choices,
value=q.client["experiment/start/cfg_sub"],
trigger=True,
)
]
else:
q.client["experiment/start/cfg_sub"] = ""
# add model type to cfg file name here
q.client["experiment/start/cfg_file"] = add_model_type(
q.client["experiment/start/cfg_file"], q.client["experiment/start/cfg_sub"]
)
if (
show_update_warnings
and q.client["experiment/start/cfg_file_prev"]
!= q.client["experiment/start/cfg_file"]
and q.client["experiment/start/cfg_category"] != "experiment"
):
items += [
ui.message_bar(type="warning", text=warning_message.format("Problem Type"))
]
show_update_warnings = False
if q.client["experiment/start/cfg_category"] == "experiment":
items += [
ui.dropdown(
name="experiment/start/cfg_experiment",
label="Experiment",
required=True,
choices=[
ui.choice(str(row.id), row["name"])
for _, row in df_experiments.iterrows()
],
value=q.client["experiment/start/cfg_experiment"],
trigger=True,
)
]
if (
show_update_warnings
and q.client["experiment/start/cfg_experiment_prev"]
!= q.client["experiment/start/cfg_experiment"]
):
items += [
ui.message_bar(
type="warning", text=warning_message.format("previous Experiment")
)
]
# Show pretrained weights toggle only for successfully finished experiments
if (
df_experiments.loc[
df_experiments.id == int(q.client["experiment/start/cfg_experiment"]),
"status",
].values[0]
== "finished"
):
items += [
ui.toggle(
name="experiment/start/cfg_experiment_pretrained",
label="Use previous experiment weights",
value=q.client["experiment/start/cfg_experiment_pretrained"],
trigger=True,
)
]
# only show yaml option, when not starting from another experiment
if q.client["experiment/start/cfg_category"] != "experiment":
items += [
ui.toggle(
name="experiment/start/from_yaml",
label="Import config from YAML",
value=False,
trigger=True,
tooltip=tooltips["experiments_import_config_from_yaml"],
)
]
if q.args["experiment/start/from_yaml"]:
items += [
ui.file_upload(
name="experiment/upload_yaml",
label="Upload!",
multiple=False,
file_extensions=["yaml"],
)
]
if q.args["experiment/upload_yaml"] is not None:
# reset previous, so the UI will be reloaded
q.client["experiment/start/cfg_file_prev"] = None
await config_import_uploaded_file(q)
logger.info(
f"PREV {q.client['experiment/start/cfg_file_prev']} "
f"{q.client['experiment/start/cfg_file']} "
f"{q.client['experiment/start/dataset_prev']} "
f"{q.client['experiment/start/dataset']} "
f"{q.client['experiment/start/cfg_experiment_prev']} "
f"{q.client['experiment/start/cfg_experiment']} "
)
# set mode to training
q.client["experiment/start/cfg_mode/mode"] = "train"
if q.client["experiment/start/cfg_category"] == "experiment":
logger.info("Starting from experiment")
# reset previous config file
q.client["experiment/start/cfg_file_prev"] = None
q.client["experiment/start/experiment"] = q.client.app_db.get_experiment(
q.client["experiment/start/cfg_experiment"]
)
parent_path = os.path.dirname(q.client["experiment/start/experiment"].path)
parent_exp_name = parent_path.split("/")[-1]
parent_experiment = f"{parent_exp_name}"
old_config = load_config_yaml(f"{parent_path}/cfg.yaml")
old_config._parent_experiment = parent_experiment
q.client["experiment/start/cfg"] = old_config
# set pretrained weights
if q.client["experiment/start/cfg_experiment_pretrained"]:
prev_weights = os.path.join(
q.client["experiment/start/experiment"].path,
"checkpoint.pth",
)
if os.path.exists(prev_weights):
q.client["experiment/start/cfg"].architecture.pretrained_weights = (
prev_weights
)
q.client["experiment/start/cfg"].architecture._visibility[
"pretrained_weights"
] = -1
experiments_df = q.client.app_db.get_experiments_df()
output_dir = os.path.abspath(
os.path.join(q.client["experiment/start/cfg"].output_directory, "..")
)
q.client["experiment/start/cfg"].experiment_name = get_unique_name(
q.client["experiment/start/cfg"].experiment_name,
experiments_df["name"].values,
lambda x: os.path.exists(os.path.join(output_dir, x)),
)
# Configuration flags:
# from_dataset -- take the values from the dataset config
# from_cfg -- take the values from the configuration file
# from_default -- take the values from the the default settings
# from_dataset_args -- take the values from the dataset's q.args
# Otherwise -- take the values from the q.args (user input)
# pick default values from config
if (
q.client["experiment/start/cfg_experiment_prev"]
!= q.client["experiment/start/cfg_experiment"]
):
q.client["experiment/start/cfg_mode/from_dataset"] = False
q.client["experiment/start/cfg_mode/from_cfg"] = True
q.client["experiment/start/cfg_mode/from_dataset_args"] = False
q.client["experiment/start/dataset"] = str(
q.client["experiment/start/experiment"].dataset
)
items[1].dropdown.value = q.client["experiment/start/dataset"]
# pick default values from config or dataset
elif (
q.client["experiment/start/dataset_prev"]
!= q.client["experiment/start/dataset"]
):
q.client["experiment/start/cfg_mode/from_dataset"] = True
q.client["experiment/start/cfg_mode/from_cfg"] = True
q.client["experiment/start/cfg_mode/from_dataset_args"] = False
# pick default values from args
else:
q.client["experiment/start/cfg_mode/from_dataset"] = False
q.client["experiment/start/cfg_mode/from_cfg"] = False
q.client["experiment/start/cfg_mode/from_dataset_args"] = True
q.client["experiment/start/cfg_mode/from_default"] = False
q.client["experiment/start/cfg_experiment_prev"] = q.client[
"experiment/start/cfg_experiment"
]
else:
logger.info("Starting from CFG")
# reset previous experiment
q.client["experiment/start/cfg_experiment_prev"] = None
# pick default values from dataset or config
if (
q.client["experiment/start/cfg_file_prev"]
!= q.client["experiment/start/cfg_file"]
) or (
q.client["experiment/start/dataset_prev"]
!= q.client["experiment/start/dataset"]
):
q.client["experiment/start/cfg_mode/from_dataset"] = True
q.client["experiment/start/cfg_mode/from_cfg"] = True
q.client["experiment/start/cfg_mode/from_default"] = True
q.client["experiment/start/cfg_mode/from_dataset_args"] = False
# pick default values from args
else:
q.client["experiment/start/cfg_mode/from_dataset"] = False
q.client["experiment/start/cfg_mode/from_cfg"] = False
q.client["experiment/start/cfg_mode/from_default"] = False
q.client["experiment/start/cfg_mode/from_dataset_args"] = True
q.client["experiment/start/cfg_file_prev"] = q.client[
"experiment/start/cfg_file"
]
config_path = (
f"llm_studio/python_configs/{q.client['experiment/start/cfg_file']}"
)
q.client["experiment/start/cfg"] = load_config_py(
config_path=config_path, config_name="ConfigProblemBase"
)
q.client["experiment/start/dataset_prev"] = q.client["experiment/start/dataset"]
logger.info(f"From dataset {q.client['experiment/start/cfg_mode/from_dataset']}")
logger.info(f"From cfg {q.client['experiment/start/cfg_mode/from_cfg']}")
logger.info(f"From default {q.client['experiment/start/cfg_mode/from_default']}")
logger.info(f"Config file: {q.client['experiment/start/cfg_file']}")
option_items = get_ui_elements(cfg=q.client["experiment/start/cfg"], q=q)
items.extend(option_items)
if q.client["experiment/start/cfg_mode/from_cfg"]:
q.page["experiment/start"] = ui.form_card(box="content", items=items)
else:
q.page["experiment/start"].items = items
q.client.delete_cards.add("experiment/start")
q.page["experiment/start/footer"] = ui.form_card(
box="footer",
items=[
ui.inline(
items=[
ui.button(
name="experiment/start/run",
label="Run experiment",
primary=True,
)
],
justify="start",
)
],
)
q.client.delete_cards.add("experiment/start/footer")
async def experiment_run(q: Q, pre: str = "experiment/start"):
"""Start an experiment.
Args:
q: Q
pre: prefix for client key
"""
# import here to avoid circular imports
from llm_studio.app_utils.sections.project import list_current_experiments
logger.info("Starting experiment")
logger.info(f"{pre}/cfg_file")
logger.info(f"CFG: {q.client[f'{pre}/cfg_file']}")
if q.client[f"{pre}/cfg_category"] == "experiment":
q.client[f"{pre}/cfg_file"] = q.client[f"{pre}/experiment"].config_file
cfg = q.client[f"{pre}/cfg"]
cfg = parse_ui_elements(cfg=cfg, q=q, pre=f"{pre}/cfg/")
cfg.experiment_name = cfg.experiment_name.replace("/", "-")
errors = check_config_for_errors(cfg)
if errors["title"] and not q.args["experiment/start/error/proceed"]:
title = (
errors["title"][0]
if len(errors["title"]) == 1
else "The following configuration mismatches were found:"
)
error_text = [ui.text(message) for message in errors["message"]]
q.page["meta"].dialog = ui.dialog(
title=title,
name="experiment/start/error/dialog",
items=error_text
+ [
ui.buttons(
[
ui.button(
name="experiment/start/error/ok", label="Ok", primary=True
),
ui.button(
name="experiment/start/error/proceed",
label="I want to proceed anyhow",
primary=False,
),
]
)
],
closable=True,
)
q.client["keep_meta"] = True
else:
start_experiment(cfg=cfg, q=q, pre=pre)
await list_current_experiments(q)
def get_experiment_table(
q, df_viz, predictions, height="calc(100vh - 245px)", actions=None
):
col_remove = [
"id",
"path",
"mode",
"seed",
"process_id",
"gpu_list",
"loss",
"eta",
"epoch",
"config_file",
]
if predictions:
col_remove += ["epoch", "val metric"]
for col in col_remove:
if col in df_viz:
del df_viz[col]
# df_viz = df_viz.rename(
# columns={"process_id": "pid", "config_file": "problem type"},
# )
# df_viz["problem type"] = df_viz["problem type"].str.replace("Text ", "")
if actions == "experiment" and q.client["experiment/list/mode"] == "train":
actions_dict = {
"experiment/list/new": "New experiment",
"experiment/list/rename": "Rename experiment",
"experiment/list/stop/table": "Stop experiment",
"experiment/list/delete/table/dialog": "Delete experiment",
}
else:
actions_dict = {}
min_widths = {
"name": "350",
"dataset": "150",
# "problem type": "190",
"metric": "75",
"val metric": "102",
"progress": "85",
"status": "90",
"info": "115",
"actions": "5" if predictions else "5",
}
if predictions:
for k, v in min_widths.items():
min_widths[k] = str(int(np.ceil(int(v) * 1.05)))
return ui_table_from_df(
q=q,
df=df_viz,
name="experiment/list/table",
sortables=["val metric"],
filterables=["name", "dataset", "problem type", "metric", "status"],
searchables=["name", "dataset"],
numerics=["val metric"],
tags=["status"],
progresses=["progress"],
min_widths=min_widths,
link_col="name",
height=height,
actions=actions_dict,
)
async def experiment_list(
q: Q,
reset: bool = True,
allowed_statuses: Optional[List[str]] = None,
actions: bool = True,
) -> None:
"""List all experiments."""
if q.client["experiment/list/mode"] is None:
q.client["experiment/list/mode"] = "train"
if q.client["experiment/list/mode"] == "train":
q.client["nav/active"] = "experiment/list"
else:
q.client["nav/active"] = "experiment/list_predictions"
if reset:
await clean_dashboard(q, mode="full")
q.client["experiment/list/df_experiments"] = get_experiments(
q,
mode=q.client["experiment/list/mode"],
status=allowed_statuses,
)
df_viz = q.client["experiment/list/df_experiments"].copy()
table = get_experiment_table(
q,
df_viz,
q.client["experiment/list/mode"] == "predict",
actions="experiment" if actions else None,
)
message_bar = get_experiment_list_message_bar(q)
items = [table, message_bar]
q.page["experiment/list"] = ui.form_card(box="content", items=items)
q.client.delete_cards.add("experiment/list")
buttons = [
ui.button(name="experiment/list/refresh", label="Refresh", primary=True),
ui.button(
name="experiment/list/compare",
label="Compare experiments",
primary=False,
),
ui.button(name="experiment/list/stop", label="Stop experiments", primary=False),
ui.button(
name="experiment/list/delete", label="Delete experiments", primary=False
),
]
q.page["dataset/display/footer"] = ui.form_card(
box="footer", items=[ui.inline(items=buttons, justify="start")]
)
q.client.delete_cards.add("dataset/display/footer")
def get_table_and_message_item_indices(q):
table_item_idx, message_item_idx = 0, 1
return table_item_idx, message_item_idx
async def experiment_compare(q: Q, selected_rows: list):
if q.client["experiment/compare/tab"] is None:
q.client["experiment/compare/tab"] = "experiment/compare/charts"
if q.args["experiment/compare/charts"] is not None:
q.client["experiment/compare/tab"] = "experiment/compare/charts"
if q.args["experiment/compare/config"] is not None:
q.client["experiment/compare/tab"] = "experiment/compare/config"
experiment_ids = [
q.client["experiment/list/df_experiments"]["id"].iloc[int(idx)]
for idx in selected_rows
]
await clean_dashboard(q, mode=q.client["experiment/compare/tab"])
tabs = [
ui.tab(name="experiment/compare/charts", label="Charts"),
ui.tab(name="experiment/compare/config", label="Config"),
]
q.page["experiment/compare/tab"] = ui.tab_card(
box="nav2", link=True, items=tabs, value=q.client["experiment/compare/tab"]
)
q.client.delete_cards.add("experiment/compare/tab")
if q.client["experiment/compare/tab"] == "experiment/compare/charts":
charts = []
experiment_names = []
for experiment_id in experiment_ids:
experiment = q.client.app_db.get_experiment(experiment_id)
experiment_path = experiment.path
charts.append(load_charts(experiment_path))
current_name = f" {experiment.name}"
experiment_names.append(current_name)
await charts_tab(q, charts, experiment_names)
elif q.client["experiment/compare/tab"] == "experiment/compare/config":
if q.client["experiment/compare/diff_toggle"] is None:
q.client["experiment/compare/diff_toggle"] = False
settings = pd.DataFrame()
for experiment_id in experiment_ids:
experiment = q.client.app_db.get_experiment(experiment_id)
experiment_path = experiment.path
experiment_cfg = load_config_yaml(os.path.join(experiment_path, "cfg.yaml"))
items = get_cfg_list_items(experiment_cfg)
act_df = pd.Series({item.label: item.value for item in items})
settings[experiment.name] = act_df
settings.index.name = "setting"
if q.client["experiment/compare/diff_toggle"]:
val_counts = settings.T.nunique()
drop_idx = val_counts[val_counts == 1].index.values
settings = settings.drop(drop_idx)
items = [
ui.toggle(
name="experiment/compare/diff_toggle",
label="Show differences only",
value=q.client["experiment/compare/diff_toggle"],
trigger=True,
),
ui_table_from_df(
q=q,
df=settings.reset_index(),
name="experiment/compare/summary/table",
link_col="setting",
height="calc(100vh - 315px)",
),
]
q.page["experiment/compare/config"] = ui.form_card(box="first", items=items)
q.client.delete_cards.add("experiment/compare/config")
buttons = [
ui.button(name="experiment/compare", label="Refresh", primary=True),
ui.button(name="experiment/list/current", label="Back", primary=False),
]
q.page["experiment/compare/footer"] = ui.form_card(
box="footer", items=[ui.inline(items=buttons, justify="start")]
)
q.client.delete_cards.add("experiment/compare/footer")
async def experiment_rename_form(q: Q, error: str = "") -> None:
experiment = q.client.app_db.get_experiment(q.client["experiment/rename/id"])
experiment_name = experiment.name
items = [
ui.textbox(
name="experiment/rename/name",
label=f"New name for {experiment_name}",
value=experiment_name,
required=True,
)
]
if error:
items.append(ui.message_bar(type="error", text=error))
q.page["experiment/list"].items = items
buttons = [
ui.button(name="experiment/rename/action", label="Rename", primary=True),
ui.button(name="experiment/list/current", label="Abort", primary=False),
]
q.page["dataset/display/footer"] = ui.form_card(
box="footer", items=[ui.inline(items=buttons, justify="start")]
)
q.client.delete_cards.add("dataset/display/footer")
async def experiment_rename_ui_workflow(q: Q):
selected_row = q.args["experiment/list/rename"]
rename_id = q.client["experiment/list/df_experiments"]["id"].iloc[int(selected_row)]
q.client["experiment/rename/id"] = rename_id
await experiment_rename_form(q)
async def experiment_rename_action(q, experiment, new_name):
"""Rename experiment with `current_id` id in DB to `new_name`"""
old_name = experiment.name
old_path = experiment.path
new_path = old_path.replace(old_name, new_name)
if old_path != new_path:
old_exp_path = f"{old_path}"
exp_path = f"{new_path}"
logger.info(f"Renaming {old_exp_path} to {exp_path}")
shutil.move(os.path.abspath(old_exp_path), os.path.abspath(exp_path))
# update the experiment name in the DB
with SqliteDict(os.path.join(new_path, "charts.db")) as charts:
for k1 in PLOT_ENCODINGS:
if k1 == "df":
# this is required to properly overwrite it
df = charts[k1].copy()
for k2, v2 in df.items():
logger.info(
f"Renaming charts {v2} to {v2.replace(old_name, new_name)}"
)
df[k2] = v2.replace(old_name, new_name)
charts[k1] = df
charts.commit()
for config_file in ["cfg.yaml"]:
config_path = os.path.join(exp_path, config_file)
if os.path.exists(config_path):
experiment_cfg = load_config_yaml(config_path)
experiment_cfg.experiment_name = new_name
experiment_cfg.output_directory = new_path
save_config_yaml(config_path, experiment_cfg)
rename_files = ["preds"]
for file in rename_files:
old_file = get_artifact_path_path(old_name, exp_path, file)
new_file = get_artifact_path_path(new_name, exp_path, file)
if os.path.exists(old_file):
logger.info(f"Renaming {old_file} to {new_file}")
shutil.move(os.path.abspath(old_file), os.path.abspath(new_file))
delete_files = ["logs"] # will be generated on demand with updates
for file in delete_files:
file = get_artifact_path_path(old_name, exp_path, file)
if os.path.exists(file):
logger.info(f"Deleting {file}")
os.remove(file)
q.client.app_db.rename_experiment(experiment.id, new_name, new_path)
async def experiment_delete(q: Q, experiment_ids: List[int]) -> None:
"""Delete selected experiments.
Args:
q: Q
experiment_ids: list of experiment ids to delete
"""
for experiment_id in experiment_ids:
experiment = q.client.app_db.get_experiment(experiment_id)
q.client.app_db.delete_experiment(experiment.id)
shutil.rmtree(f"{experiment.path}")
async def experiment_stop(q: Q, experiment_ids: List[int]) -> None:
"""Stop selected experiments.
Args:
q: Q
experiment_ids: list of experiment ids to stop
"""
for experiment_id in experiment_ids:
experiment = q.client.app_db.get_experiment(experiment_id)
try:
ret = kill_child_processes(int(experiment.process_id))
if ret:
flag_path = os.path.join(experiment.path, "flags.json")
write_flag(flag_path, "status", "stopped")
except Exception as e:
logger.error(f"Error while stopping the experiment: {e}")
pass
def load_charts(experiment_path):
try:
with SqliteDict(os.path.join(experiment_path, "charts.db")) as charts:
charts = dict(charts)
except Exception:
charts = {}
logger.warning("Too early, wait for the charts to appear")
return charts
async def experiment_display(q: Q) -> None:
"""Display a selected experiment."""
experiment_id = q.client["experiment/list/df_experiments"]["id"].iloc[
q.client["experiment/display/id"]
]
q.client["experiment/display/experiment_id"] = experiment_id
experiment = q.client.app_db.get_experiment(experiment_id)
q.client["experiment/display/experiment"] = experiment
q.client["experiment/display/experiment_path"] = experiment.path
status, _ = get_experiment_status(experiment.path)
charts = load_charts(q.client["experiment/display/experiment_path"])
q.client["experiment/display/charts"] = charts
if experiment.mode == "train":
if q.client["experiment/display/tab"] is None:
q.client["experiment/display/tab"] = "experiment/display/charts"
else:
if q.client["experiment/display/tab"] is None:
q.client["experiment/display/tab"] = "experiment/display/summary"
if q.args["experiment/display/charts"] is not None:
q.client["experiment/display/tab"] = "experiment/display/charts"
if q.args["experiment/display/summary"] is not None:
q.client["experiment/display/tab"] = "experiment/display/summary"
if q.args["experiment/display/train_data_insights"] is not None:
q.client["experiment/display/tab"] = "experiment/display/train_data_insights"
if q.args["experiment/display/validation_prediction_insights"] is not None:
q.client["experiment/display/tab"] = (
"experiment/display/validation_prediction_insights"
)
if q.args["experiment/display/config"] is not None:
q.client["experiment/display/tab"] = "experiment/display/config"
if q.args["experiment/display/deployment"] is not None:
q.client["experiment/display/tab"] = "experiment/display/deployment"
if q.args["experiment/display/logs"] is not None:
q.client["experiment/display/tab"] = "experiment/display/logs"
if q.args["experiment/display/chat"] is not None:
q.client["experiment/display/tab"] = "experiment/display/chat"
await clean_dashboard(q, mode=q.client["experiment/display/tab"])
tabs = [
ui.tab(name="experiment/display/charts", label="Charts"),
ui.tab(name="experiment/display/summary", label="Summary"),
]
# html for legacy experiments
has_train_data_insights = any(
[
charts.get(plot_encoding, dict()).get("train_data") is not None
for plot_encoding in PLOT_ENCODINGS
]
)
if has_train_data_insights:
tabs += [
ui.tab(
name="experiment/display/train_data_insights",
label="Train Data Insights",
)
]
has_validation_prediction_insights = any(
[
charts.get(plot_encoding, dict()).get("validation_predictions") is not None
for plot_encoding in PLOT_ENCODINGS
]
)
if has_validation_prediction_insights:
tabs += [
ui.tab(
name="experiment/display/validation_prediction_insights",
label="Validation Prediction Insights",
)
]
tabs += [
ui.tab(name="experiment/display/logs", label="Logs"),
ui.tab(name="experiment/display/config", label="Config"),
]
if status == "finished":
tabs += [ui.tab(name="experiment/display/chat", label="Chat")]
q.page["experiment/display/tab"] = ui.tab_card(
box="nav2", link=True, items=tabs, value=q.client["experiment/display/tab"]
)
q.client.delete_cards.add("experiment/display/tab")
if q.client["experiment/display/tab"] == "experiment/display/charts":
await charts_tab(q, [charts], [""])
elif q.client["experiment/display/tab"] in [
"experiment/display/train_data_insights",
"experiment/display/validation_prediction_insights",
]:
await insights_tab(charts, q)
elif q.client["experiment/display/tab"] in ["experiment/display/summary"]:
await summary_tab(experiment_id, q)
elif q.client["experiment/display/tab"] in ["experiment/display/config"]:
await configs_tab(q)
elif q.client["experiment/display/tab"] in ["experiment/display/logs"]:
await logs_tab(q)
elif q.client["experiment/display/tab"] in ["experiment/display/chat"]:
await chat_tab(q)
await q.page.save()
buttons = [
ui.button(name="experiment/display/refresh", label="Refresh", primary=True)
]
buttons += [
ui.button(
name="experiment/display/download_logs",
label="Download logs/config",
primary=False,
)
]
if status == "finished":
buttons += [
ui.button(
name="experiment/display/download_predictions",
label="Download predictions",
primary=False,
disabled=False,
tooltip=None,
),
ui.button(
name="experiment/display/download_model",
label="Download model",
primary=False,
disabled=False,
tooltip=None,
),
ui.button(
name="experiment/display/push_to_huggingface",
label="Push checkpoint to huggingface",
primary=False,
disabled=False,
tooltip=None,
),
]
buttons += [ui.button(name="experiment/list/current", label="Back", primary=False)]
q.page["experiment/display/footer"] = ui.form_card(
box="footer",
items=[
ui.inline(items=buttons, justify="start"),
],
)
q.client.delete_cards.add("experiment/display/footer")
async def insights_tab(charts, q):
if q.client["experiment/display/tab"] == "experiment/display/train_data_insights":
key = "train_data"
elif (
q.client["experiment/display/tab"]
== "experiment/display/validation_prediction_insights"
):
key = "validation_predictions"
for k1 in PLOT_ENCODINGS:
if k1 not in charts:
continue
for k2, v2 in charts[k1].items():
if k2 != key:
continue
if k1 == "html":
q.page[f"experiment/display/charts/{k1}_{k2}"] = ui.markup_card(
box="first", title="", content=v2
)
q.client.delete_cards.add(f"experiment/display/charts/{k1}_{k2}")
continue
elif k1 == "image":
q.page[f"experiment/display/charts/{k1}_{k2}"] = ui.image_card(
box="first", title="", type="png", image=v2
)
q.client.delete_cards.add(f"experiment/display/charts/{k1}_{k2}")
continue
elif k1 == "df":
df = pd.read_parquet(v2)
min_widths = {
col: "350" for col in df.columns if "text" in str(col).lower()
}
#
if key == "train_data":
min_widths["Content"] = "800"
q.page[f"experiment/display/charts/{k1}_{k2}"] = ui.form_card(
box="first",
items=[
ui_table_from_df(
q=q,
df=df,
name=f"experiment/display/charts/{k1}_{k2}",
sortables=[
col for col in df.columns if col.startswith("Metric")
],
markdown_cells=[
col
for col in df.columns
if not col.startswith("Metric")
],
searchables=list(df.columns),
downloadable=True,
resettable=True,
min_widths=min_widths,
height="calc(100vh - 245px)",
max_char_length=50_000,
cell_overflow="tooltip",
)
],
)
q.client.delete_cards.add(f"experiment/display/charts/{k1}_{k2}")
continue
async def summary_tab(experiment_id, q):
experiment_df = get_experiments(q)
input_dict = experiment_df[experiment_df.id == experiment_id].iloc[0].to_dict()
cfg = load_config_yaml(
os.path.join(q.client["experiment/display/experiment_path"], "cfg.yaml")
)
_ = get_tokenizer(cfg)
# experiment card
card_name = "experiment/display/summary/experiment"
q.page[card_name] = ui.form_card(
box=ui.box(zone="first"),
items=[
ui.separator("Experiment"),
ui.stats(
[
ui.stat(
value=cfg.experiment_name,
label="Name",
),
],
justify="between",
inset=True,
),
ui.stats(
[
ui.stat(
value=input_dict["config_file"],
label="Problem Type",
),
],
justify="between",
inset=True,
),
],
)
q.client.delete_cards.add(card_name)
# datasets card
card_name = "experiment/display/summary/datasets"
q.page[card_name] = ui.form_card(
box=ui.box(zone="first"),
items=[
ui.separator("Datasets"),
ui.stats(
[
ui.stat(
value=Path(cfg.dataset.train_dataframe).stem,
label="Training Dataset",
),
],
justify="between",
inset=True,
),
ui.stats(
[
ui.stat(
value=(
"-"
if cfg.dataset.validation_dataframe in ["", "None", None]
else Path(cfg.dataset.validation_dataframe).stem
),
label="Validation Dataset",
),
],
justify="between",
inset=True,
),
],
)
q.client.delete_cards.add(card_name)
# score card
card_name = "experiment/display/summary/score"
q.page[card_name] = ui.form_card(
box=ui.box(zone="first"),
items=[
ui.separator("Score"),
ui.stats(
[
ui.stat(
value=input_dict["metric"],
label="Metric",
),
],
justify="between",
inset=True,
),
ui.stats(
[
ui.stat(
value=(
"-"
if input_dict["val metric"] in ["", "None", None]
else str(input_dict["val metric"])
),
label="Validation Score",
),
],
justify="between",
inset=True,
),
],
)
q.client.delete_cards.add(card_name)
# main configs card
card_name = "experiment/display/summary/main_configs"
q.page[card_name] = ui.form_card(
box=ui.box(zone="second"),
items=[
ui.separator("Main Configurations"),
ui.stats(
[
ui.stat(
value=cfg.llm_backbone,
label="LLM Backbone",
),
ui.stat(
value=str(cfg.training.lora),
label="Lora",
),
ui.stat(
value=str(cfg.training.epochs),
label="Epochs",
),
ui.stat(
value=str(cfg.training.batch_size),
label="Batch Size",
),
],
justify="between",
inset=True,
),
ui.stats(
[
ui.stat(
value=str(input_dict["loss"]),
label="Loss Function",
),
ui.stat(
value=cfg.architecture.backbone_dtype,
label="Backbone Dtype",
),
ui.stat(
value=str(cfg.architecture.gradient_checkpointing),
label="Gradient Checkpointing",
),
ui.stat(
value=input_dict["gpu_list"],
label="GPU List",
),
],
justify="between",
inset=True,
),
],
)
q.client.delete_cards.add(card_name)
# code card
card_name = "experiment/display/summary/code"
content = get_experiment_summary_code_card(cfg=cfg)
q.page[card_name] = ui.markdown_card(
box=ui.box(zone="third"),
title="",
content=content,
)
q.client.delete_cards.add(card_name)
async def configs_tab(q):
experiment_cfg = load_config_yaml(
os.path.join(q.client["experiment/display/experiment_path"], "cfg.yaml")
)
items = get_cfg_list_items(experiment_cfg)
q.page["experiment/display/config"] = ui.stat_list_card(
box="first", items=items, title=""
)
q.client.delete_cards.add("experiment/display/config")
async def logs_tab(q):
logs_path = f"{q.client['experiment/display/experiment_path']}/logs.log"
text = ""
in_pre = 0
# Read log file only if it already exists
if os.path.exists(logs_path):
with open(logs_path, "r") as f:
for line in f.readlines():
if in_pre == 0:
text += "<div>"
if "INFO: Lock" in line:
continue
# maximum line length
n = 250
chunks = [line[i : i + n] for i in range(0, len(line), n)]
text += "</div><div>".join(chunks)
# Check for formatted HTML text
if "<pre>" in line:
in_pre += 1
if "</pre>" in line:
in_pre -= 1
if in_pre == 0:
text += "</div>"
items = [ui.text(text)]
q.page["experiment/display/logs"] = ui.form_card(box="first", items=items, title="")
q.client.delete_cards.add("experiment/display/logs")
def subsample(key1, key2, value, max_plot_points=1000):
act_plot_points = len(value["steps"])
if act_plot_points > max_plot_points:
stride = int(np.ceil(act_plot_points / max_plot_points))
value["steps"] = value["steps"][::stride]
value["values"] = value["values"][::stride]
logger.info(
f"{key1} {key2} sampled from size {act_plot_points} to size "
f"{len(value['steps'])} using stride {stride}."
)
return value
def unite_validation_metric_charts(charts_list):
unique_metrics = []
for chart in charts_list:
unique_metrics.extend(list(chart.get("validation", {}).keys()))
unique_metrics = set([key for key in unique_metrics if key != "loss"])
if len(unique_metrics) > 1:
for chart in charts_list:
if "validation" in chart:
for key in unique_metrics:
if key in chart["validation"]:
chart["validation"]["metric"] = chart["validation"][key]
del chart["validation"][key]
return charts_list
async def charts_tab(q, charts_list, legend_labels):
charts_list = unite_validation_metric_charts(charts_list)
box = ["first", "first", "second", "second"]
cnt = 0
for k1 in ["meta", "train", "validation"]:
if all([k1 not in charts for charts in charts_list]):
continue
all_second_keys: Set = set()
for charts in charts_list:
if k1 in charts:
all_second_keys = all_second_keys.union(set(charts[k1].keys()))
# Always plot loss in the lower left corner
if "loss" in all_second_keys:
all_second_keys.remove("loss")
list_all_second_keys = ["loss"] + list(all_second_keys)
else:
list_all_second_keys = list(all_second_keys)
for k2 in list_all_second_keys:
logger.info(f"{k1} {k2}")
items = []
tooltip = ""
if k1 == "meta" and k2 == "lr":
tooltip = "Current learning rate throughout the training process."
elif k1 == "train" and k2 == "loss":
tooltip = (
"Current training loss throughout the training process. "
"Loss is calculated as the average of the last ten batches."
)
elif k1 == "validation" and k2 == "loss":
tooltip = (
"Current validation loss throughout the training process. "
"Loss is calculated as the average of all validation batches. "
)
elif k1 == "validation" and k2 != "loss":
tooltip = (
"Current validation metric throughout the training process. "
"Metric is calculated on full validation set predictions."
)
else:
continue
title = f"{k1} {k2}".upper().replace("META LR", "LEARNING RATE")
if k2 == "loss":
title = title.replace("LOSS", "BATCH LOSS")
items.append(ui.text(title, tooltip=tooltip))
rows = []
max_samples = q.client["chart_plot_max_points"]
for charts, label in zip(charts_list, legend_labels):
if k1 not in charts or k2 not in charts[k1]:
continue
v2 = charts[k1][k2]
v2 = subsample(k1, k2, v2, max_samples)
if k2 == "lr" and "lr_diff" in charts["meta"]:
v3 = charts["meta"]["lr_diff"]
v3 = subsample("meta", "lr_diff", v3, max_samples)
rows.extend(
[
(v2["steps"][i], f"learning rate{label}", v2["values"][i])
for i in range(len(v2["values"]))
]
+ [
(
v3["steps"][i],
f"differential learning rate{label}",
v3["values"][i],
)
for i in range(len(v3["values"]))
]
)
color = "=type"
fields = ["step", "type", "value"]
elif len(charts_list) > 1:
rows.extend(
[
(v2["steps"][i], label.strip(), v2["values"][i])
for i in range(len(v2["values"]))
]
)
color = "=type"
fields = ["step", "type", "value"]
else:
rows.extend(
[
(v2["steps"][i], v2["values"][i]) # type: ignore
for i in range(len(v2["values"]))
]
)
color = wave_theme.color
fields = ["step", "value"]
d = data(fields=fields, rows=rows, pack=True)
viz = ui.visualization(
plot=ui.plot(
[
ui.mark(
type="line",
x_title="step",
x_scale="linear",
y_scale="linear",
x="=step",
y="=value",
color=color,
y_min=0 if k1 == "meta" and k2 == "lr" else None,
color_range=wave_theme.color_range,
)
]
),
data=d, # type: ignore
interactions=["brush"],
height="calc((100vh - 275px)*0.41)",
width="560px",
)
items.append(viz)
if k1 == "validation" and k2 == "loss" and np.sum(v2["values"]) == 0:
items.append(
ui.message_bar(
type="info",
text="Validation batch loss cannot be \
calculated for this problem type.",
)
)
q.page[f"experiment/display/charts/{k1}_{k2}"] = ui.form_card(
box=box[cnt], items=items
)
q.client.delete_cards.add(f"experiment/display/charts/{k1}_{k2}")
cnt += 1
async def experiment_artifact_build_error_dialog(q: Q, error: str):
q.page["meta"].dialog = ui.dialog(
"Failed to build artifact", items=[ui.text(error)], closable=True
)
q.client["keep_meta"] = True
async def experiment_download_artifact(
q: Q,
get_artifact_path_fn: Callable[[str, str], str],
save_artifact_fn: Callable[[str, str], str],
additional_log: Optional[str] = "",
min_disk_space: Optional[float] = 0.0,
):
"""Download specific artifact, if it does not exist, create it on demand
Args:
q: Q
get_artifact_path_fn: function that returns path to the artifact
save_artifact_fn: function that generates the artifact and returns its path
additional_log: additional information to be logged
min_disk_space: minimal disk available needed to generate artifact
"""
experiment = q.client["experiment/display/experiment"]
experiment_path = q.client["experiment/display/experiment_path"]
zip_path = get_artifact_path_fn(experiment.name, experiment_path)
if not os.path.exists(zip_path):
try:
check_available_space(experiment_path, min_disk_space)
except LLMResourceException as e:
error = f"Cannot create {os.path.basename(zip_path)}. {e}"
await experiment_artifact_build_error_dialog(q, error)
return
logger.info(f"Creating {zip_path} on demand")
zip_path = save_artifact_fn(experiment.name, experiment_path)
if additional_log:
logger.info(f"{additional_log}: {zip_path}")
q.page["meta"].script = ui.inline_script(
f'window.open("{get_download_link(q, zip_path)}", "_blank");'
)
await q.page.save()
async def experiment_download_predictions(q: Q):
"""Download experiment predictions."""
await experiment_download_artifact(
q, get_predictions_path, save_prediction_outputs, "Predictions path", None
)
async def experiment_download_logs(q: Q):
"""Download experiment logs."""
experiment = q.client["experiment/display/experiment"]
experiment_path = q.client["experiment/display/experiment_path"]
zip_path = get_logs_path(experiment.name, experiment_path)
if not os.path.exists(zip_path):
logs = q.client["experiment/display/charts"]
logger.info(f"Creating {zip_path} on demand")
zip_path = save_logs(experiment.name, experiment_path, logs)
download_url = get_download_link(q, zip_path)
logger.info(f"Logs URL: {download_url}")
q.page["meta"].script = ui.inline_script(
f'window.open("{download_url}", "_blank");'
)
await q.page.save()
async def config_import_uploaded_file(q: Q):
""" "Importing a config file from drag and drop to the filesystem"""
file_url = q.args["experiment/upload_yaml"][0]
file_name = file_url.split("/")[-1]
path = f"{get_data_dir(q)}/{file_name}"
local_path = await q.site.download(file_url, path)
await q.site.unload(q.args["experiment/upload_yaml"][0])
with open(local_path, "r") as f:
yaml_data = yaml.safe_load(f)
yaml_data = flatten_dict(yaml_data)
q.client["experiment/yaml_data"] = yaml_data
async def show_message(q, msg_key, page, idx, msg_type):
info = q.client[msg_key]
if info:
q.page[page].items[idx].message_bar.text = info
q.page[page].items[idx].message_bar.type = msg_type
q.client[msg_key] = ""
def get_experiment_list_message_bar(q):
if q.client["experiment_halt_reason"]:
msg_bar = ui.message_bar(type="error", text=q.client["experiment_halt_reason"])
del q.client["experiment_halt_reason"]
elif q.client["force_disable_pipelines"]:
msg_bar = ui.message_bar(type="info", text=q.client["force_disable_pipelines"])
del q.client["force_disable_pipelines"]
else:
msg_bar = ui.message_bar(type="info", text="")
return msg_bar
async def experiment_download_model(q: Q):
experiment = q.client["experiment/display/experiment"]
experiment_path = q.client["experiment/display/experiment_path"]
zip_path = get_model_path(experiment.name, experiment_path)
if not os.path.exists(zip_path):
logger.info(f"Creating {zip_path} on demand")
cfg = load_config_yaml(os.path.join(experiment_path, "cfg.yaml"))
device = "cuda"
experiments = get_experiments(q)
num_running_queued = len(
experiments[experiments["status"].isin(["queued", "running"])]
)
if num_running_queued > 0 or (
cfg.training.lora and cfg.architecture.backbone_dtype in ("int4", "int8")
):
logger.info("Preparing model on CPU. This might slow down the progress.")
device = "cpu"
with set_env(HUGGINGFACE_TOKEN=q.client["default_huggingface_api_token"]):
cfg, model, tokenizer = load_cfg_model_tokenizer(
experiment_path, merge=True, device=device
)
model = unwrap_model(model)
checkpoint_path = cfg.output_directory
model_save_time = time.time()
model.backbone.save_pretrained(checkpoint_path)
# See PreTrainedTokenizerBase.save_pretrained for documentation
# Safeguard against None return if tokenizer class is
# not inherited from PreTrainedTokenizerBase
tokenizer_files = list(tokenizer.save_pretrained(checkpoint_path) or [])
card = get_model_card(cfg, model, repo_id="<path_to_local_folder>")
card.save(os.path.join(experiment_path, "model_card.md"))
logger.info(f"Creating Zip File at {zip_path}")
zf = zipfile.ZipFile(zip_path, "w")
FILES_TO_PUSH = [
"vocab.json",
"sentencepiece.bpe.model",
"bpe_encoder.bin",
"tokenizer_config.json",
"tokenizer.json",
"special_tokens_map.json",
"merges.txt",
"generation_config.json",
"config.json",
"added_tokens.json",
"model_card.md",
"classification_head.pth",
]
FILES_TO_PUSH = set(
FILES_TO_PUSH
+ [os.path.split(tokenizer_file)[-1] for tokenizer_file in tokenizer_files]
)
# Add tokenizer and config.json files, as well as potential classification head
paths_added = []
for file in FILES_TO_PUSH:
path = os.path.join(experiment_path, file)
if os.path.isfile(path):
paths_added.append(path)
add_file_to_zip(zf=zf, path=path)
# Add model weight files. save_pretrained() does not return the saved files
weight_paths = glob.glob(os.path.join(checkpoint_path, "pytorch_model*.*"))
for path in weight_paths:
paths_added.append(path)
add_file_to_zip(zf=zf, path=path)
# Add all files that were created after the model was saved.
# This is useful for potential changes/different
# naming conventions across different backbones.
for file in os.listdir(checkpoint_path):
file_path = os.path.join(checkpoint_path, file)
if (
os.path.getmtime(file_path) > model_save_time
and file_path not in paths_added
and file_path != zip_path
):
add_file_to_zip(zf=zf, path=file_path)
paths_added.append(file_path)
logger.info(
f"Added {file_path} to zip file as it "
"was created when saving the model state."
)
zf.close()
download_url = get_download_link(q, zip_path)
logger.info(f"Logs URL: {download_url}")
q.page["meta"].script = ui.inline_script(
f'window.open("{download_url}", "_blank");'
)
await q.page.save()
async def experiment_push_to_huggingface_dialog(q: Q, error: str = ""):
if q.args["experiment/display/push_to_huggingface"] or error:
devices = ["cpu", "cpu_shard"] + [
f"cuda:{idx}" for idx in range(torch.cuda.device_count())
]
default_device = "cuda:0"
experiments = get_experiments(q)
num_running_queued = len(
experiments[experiments["status"].isin(["queued", "running"])]
)
experiment_path = q.client["experiment/display/experiment_path"]
cfg = load_config_yaml(os.path.join(experiment_path, "cfg.yaml"))
if num_running_queued > 0 or cfg.environment.use_deepspeed:
default_device = "cpu"
try:
huggingface_hub.login(q.client["default_huggingface_api_token"])
user_id = huggingface_hub.whoami()["name"]
except Exception:
user_id = ""
dialog_items = [
ui.message_bar("error", error, visible=True if error else False),
ui.textbox(
name="experiment/display/push_to_huggingface/account_name",
label="Account Name",
value=user_id,
width="500px",
required=False,
tooltip=(
"The account name on HF to push the model to. "
"Leaving it empty will push it to the default user account."
),
),
ui.textbox(
name="experiment/display/push_to_huggingface/model_name",
label="Model Name",
value=hf_repo_friendly_name(
q.client["experiment/display/experiment"].name
),
width="500px",
required=True,
tooltip="The name of the model as shown on HF.",
),
ui.dropdown(
name="experiment/display/push_to_huggingface/device",
label="Device for preparing the model",
required=True,
value=default_device,
width="500px",
choices=[ui.choice(str(d), str(d)) for d in devices],
tooltip=(
"The local device to prepare the model before pushing it to HF. "
"CPU will never load the weights to the GPU, which can be useful "
"for large models, but will be significantly slower. "
"Cpu_shard will first load on CPU and then shard on all GPUs "
"before pushing to HF."
),
),
ui.textbox(
name="experiment/display/push_to_huggingface/api_key",
label="Huggingface API Key",
value=q.client["default_huggingface_api_token"],
width="500px",
password=True,
required=True,
tooltip="HF API key, needs write access.",
),
ui.toggle(
name="default_safe_serialization",
label="Use Hugging Face safetensors for safe serialization",
value=q.client["default_safe_serialization"],
),
ui.buttons(
[
ui.button(
name="experiment/display/push_to_huggingface_submit",
label="Export",
primary=True,
),
ui.button(name="cancel", label="Cancel", primary=False),
]
),
]
elif q.args["experiment/display/push_to_huggingface_submit"]:
await busy_dialog(
q=q,
title="Exporting to HuggingFace",
text="Model size can affect the export time significantly.",
)
experiment_path = q.client["experiment/display/experiment_path"]
device = q.client["experiment/display/push_to_huggingface/device"]
api_key = q.client["experiment/display/push_to_huggingface/api_key"]
user_id = q.client["experiment/display/push_to_huggingface/account_name"]
safe_serialization = q.client["default_safe_serialization"]
model_name = q.client[
"experiment/display/push_to_huggingface/model_name"
].replace(".", "-")
publish_model_to_hugging_face(
path_to_experiment=experiment_path,
device=device,
api_key=api_key,
user_id=user_id,
model_name=model_name,
safe_serialization=safe_serialization,
)
dialog_items = [
ui.message_bar("success", "Success"),
ui.buttons(
[
ui.button(name="ok", label="OK", primary=True),
]
),
]
dialog = ui.dialog(
title="Push to HuggingFace Hub",
items=dialog_items,
closable=True,
name="push_to_huggingface_dialog",
)
q.page["meta"].dialog = dialog
q.client["keep_meta"] = True
def get_experiment_summary_code_card(cfg) -> str:
repo_id: Optional[str] = None
hf_yaml_path = f"{cfg.output_directory}/hf.yaml"
with open(
os.path.join("model_cards", cfg.environment._summary_card_template), "r"
) as f:
text = f.read()
if os.path.exists(hf_yaml_path):
with open(hf_yaml_path, "r") as fp:
repo_id = yaml.load(fp, Loader=yaml.FullLoader)["repo_id"]
if repo_id is None:
repo_id = "account/model"
# Model repo
text = text.replace("{{repo_id}}", repo_id)
# Versions
text = text.replace("{{transformers_version}}", transformers.__version__)
text = text.replace("{{einops_version}}", einops.__version__)
text = text.replace("{{accelerate_version}}", accelerate.__version__)
text = text.replace("{{torch_version}}", torch.__version__)
# Configs
text = text.replace("{{text_prompt_start}}", str(cfg.dataset.text_prompt_start))
text = text.replace(
"{{text_answer_separator}}", str(cfg.dataset.text_answer_separator)
)
text = text.replace(
"{{end_of_sentence}}",
str(cfg._tokenizer_eos_token) if cfg.dataset.add_eos_token_to_prompt else "",
)
text = text.replace("{{trust_remote_code}}", str(cfg.environment.trust_remote_code))
if cfg.problem_type not in NON_GENERATION_PROBLEM_TYPES:
text = text.replace(
"{{min_new_tokens}}", str(cfg.prediction.min_length_inference)
)
text = text.replace(
"{{max_new_tokens}}", str(cfg.prediction.max_length_inference)
)
text = text.replace("{{use_fast}}", str(cfg.tokenizer.use_fast))
text = text.replace("{{do_sample}}", str(cfg.prediction.do_sample))
text = text.replace("{{num_beams}}", str(cfg.prediction.num_beams))
text = text.replace("{{temperature}}", str(cfg.prediction.temperature))
text = text.replace(
"{{repetition_penalty}}", str(cfg.prediction.repetition_penalty)
)
return text
|